Towards Causal Effect Estimation of Emotional Labeling of Watched Videos

Abstract

Emotions play a crucial role in human life, they are measured using many approaches. There are also many methodologies for emotion elicitation. Emotion elicitation through video watching is one important approach used to create emotion datasets. However, the causation link between video content and elicited emotions was not well explained by scientific research. In this article, we present an approach for computing the causal effect of video content on elicited emotion. The Do-Calculus theory was employed for computing causal inference, and a SCM (Structured Causal Model) was proposed considering the following variables: EEG signal, age, gender, video content, like/dislike, and emotional quadrant. To evaluate the approach, EEG data were collected from volunteers watching a sample of videos from the LIRIS-ACCEDE dataset. A total of 48 causal effects was statistically evaluated in order to check the causal impact of age, gender, and video content on liking and emotion. The results show that the approach can be generalized for any dataset that contains the variables of the proposed SCM. Furthermore, the proposed approach can be applied to any other similar dataset if an appropriate SCM is provided.